Per: https://github.com/joyent/node/pull/1879#issuecomment-2424882
Some concerns have been raised about the new cluster functionality built
into core.

I personally feel this kind of functionality should be special outside of
the node-core binary, but still be part of the node package.

I propose a node-app (or similar named) binary, that wraps node and provides
the extra functionality such as cluster, and leaving room for future
expansion into configuring "apps" of node instead of libraries, before we go
down the road of node --foo --bar --baz --cluster-main myApp.js
--cluster-workers worker.js

I propose we continue to expand on package.json, and have a binary you pass
a config file to to load.

For example:

./myapp

#!/usr/bin/env node-app
{
  "name": "myapp"
  "app": {
    "main": "cluster",
    "cluster": {
      "master": "app/master.js",
      "workers": {
        "app/workers/worker1.js": "",
        "app/workers/worker2.js": "1"
      }
    }
  }

What do you guys think?

and windows would be "C:\node\node-app.exe C:\MyApp\MyApp.json"

And ideally, would CWD into the dir of the .json file for execution, and let
the .json file specify a different CWD relative (using path.resolve based
off CWD of .json) to the .json file.

and if node-app is invoked without a first arg, it looks for package.json
(or maybe a diff filename if you guys feel is more appropriate) in current
directory

so cd /myApp/
node-app

I really like this idea a lot!

Brian

You wouldn't consider configuring advanced features such as cluster in a
JSON file simpler than appending magical keywords onto the binary?

Having a standardized ways of developing/configuring apps from node-core in
an already established JSON file method Id think would be ideal..

What happens when you guys consider another feature 'core worthy', so  we
make a munged special keyword that does both?

I'm just really afraid of putting keywords into the arguments (and before
the filename at that, hurting shebang simplicity) is going to limit future
progression of node.

#!/usr/bin/env node cluster
// runs in node cluster mode

I disagree that it hurts shebangs, but it does hurt app-level
configurability.

In Haraka for example I have nodes=N|cpus in a config file. If it is there
(and not zero) it runs "clustered" (I hate that term, but it is familiar for
now).

That's now really hard to do. You can't do it from within node any more.

Matt.

If you have an webserver - probably you have an configuration file, its how
things used to be.  It will be bad if node-cluster configuration and
application config will be in the different places, so the solution is the
default config provided by the node cluster and accessible from the
application.

On 19 October 2011 06:42, Matt <hel...@gmail.com> wrote: